A Statistical Link between Learning and Evolution
Widely replicated experimental findings show that faced with multiple variable reward outcomes, animals sample each proportional to their mean payoff. This finding is explained computationally using the Holland theorem, in which this style of sampling is optimal given certain assumptions about the distribution. Other adaptive phenomena which seem to be consistent with this law are discussed, and it is suggested that the ‘adaptive problem’ may be broadly similar across different domains.
KeywordsExpected Loss Slot Machine Bandit Problem Adaptive Problem Adaptive Phenomenon
Unable to display preview. Download preview PDF.
- Holland, J. (1992), Adaptation in Natural and Artificial Systems, University of Michigan Press, Ann-Arbor.Google Scholar
- Levy, W. and Desmond, N. (1985), “The Rules of Elemental Synaptic Plasticity”, from Levy, W., Anderson, J. and Lehmkuhle, S. (eds), Synaptic Modification, Neuron Selectivity, and Nervous System Organisation, Lawrence Erlbaum Associates.Google Scholar
- Thompson, R. (1993), The Brain, W.H.Freeman and Co., New York.Google Scholar